In a Worker, you always have to cache.match and return that response yourself so it’s up to you at w

In a Worker, you always have to cache.match and return that response yourself so it’s up to you at which part of the flow you want to return it

Kkian In a Worker, you always have to cache.match and return that response yourself so...

Isaac McFadyen•5/30/23, 8:05 PM

This - just don't cache protected routes. The idea of versioning is to help separate pages that you do cache, but just don't cache the authenticated routes and you should be good.

Vercantez•5/30/23, 8:23 PM

Absolutely loving Workers. Having one issue though: my paid worker's context (global variables) are resetting after only a couple of minutes of inactivity. I realize global variables are not persistent but the documentation states that nodes are "occasionally evicted from memory" which kinda implies longer than a couple minutes imo. https://developers.cloudflare.com/workers/platform/limits/#memory am I doing something wrong? Can I monitor node evictions?

Limits · Cloudflare Workers docs

You can request adjustments to limits that conflict with your project goals by contacting Cloudflare. To request an increase to a limit, complete the …

kianOP•5/30/23, 8:27 PM

I suspect it's not that it's being evicted, you just don't hit the same Worker.

kianOP•5/30/23, 8:28 PM

Isolates are per server, and Cloudflare has thousands.

Unsmart•5/30/23, 8:28 PM

Nodes can get evicted at any second pretty much unless processing a request and in the first 30 seconds of it. And no you cant monitor them. But also you are probably hitting a different server
And afaik they are working to completely remove the ability for global vars between requests but maybe I am mistaken on that part. Mightve just been the local storage thing I am thinking about

Vercantez•5/30/23, 8:42 PM

Ah ok! Thanks for the quick response. Is there a way to log a server identifier? Are Workers deployed to many servers in the same data center? The link above states "Only one Workers instance runs on each of the many global Cloudflare global network servers." and I assumed that traffic from the same IP would hit the same server. Thanks again for the help

kianOP•5/30/23, 8:44 PM

The same keep-alive connection (i.e in a browser) will have affinity with a given server but new connections will probably go to different servers.

kianOP•5/30/23, 8:45 PM

You can store a Worker 'identifier' in the global scope to identify an isolate, and otherwise do a subrequest to https://cloudflare.com/cdn-cgi/fetchhttps://cloudflare.com/cdn-cgi/fetch to get the colo/metal identifier

kianOP•5/30/23, 8:45 PM

fwiw, I've had isolates stay up for as long as 22 hours

kianOP•5/30/23, 8:45 PM

it's just rare-ish for you to hit the same isolate

Chaika•5/30/23, 8:46 PM

Workers run on every metal, If you need consistency, look into Durable Objects (increasing a counter, consistent state). If you're doing something like pushing items into a queue to be dealt with later, look into CF Queues

kianOP•5/30/23, 8:46 PM

if user.comment = "that guy who keeps running weird experiments" { ... }if user.comment = "that guy who keeps running weird experiments" { ... }

kianOP•5/30/23, 8:47 PM

they'll wonder why I have so many Array.fromArray.from making 100,000,000 random entries

Vercantez•5/30/23, 8:47 PM

Oh, yeah that explains it then! Thanks a bunch! I'm just caching KV values to avoid async IO for a latency sensitive application

Chaika•5/30/23, 8:48 PM

You could use local colo cache as well/along with (https://developers.cloudflare.com/workers/runtime-apis/cache/), there's a community made library called BetterKV that does it for you: https://flareutils.pages.dev/betterkv/
(and you don't pay for cache ops, but you do pay for KV ops even if cached)
KV is going to be pretty low latency though, except on cold reads

BetterKV | Flareutils

Small Utilities and little goodies that make developing with Cloudflare easier and faster.

Vercantez•5/30/23, 8:52 PM

Looks useful, thanks! Yeah I'm probably just prematurely optimizing

Kavatch•5/30/23, 8:55 PM

Should the identifieidentifier or the zone_identifierzone_identifier be considered private / confidential?
Like if I have them in a git repo should I make sure to remove them before committing or can I leave them?

kianOP•5/30/23, 9:01 PM

They're not private, no

kianOP•5/30/23, 9:02 PM

I mean theoretically I guess if you leaked an API token that had access to 1 zone, and not to list zones, it'd be useless (ish) without that ID but that's security through obscurity so a bad idea anyways

Kavatch•5/30/23, 9:05 PM

Alright thank you!

PPanvA **📢 New OAuth 2 / OpenID Connect Client library** 👋 author of https://github....

dave•5/31/23, 12:35 AM

This is really nice, thank you!!

dave•5/31/23, 1:14 AM

@sathoro Have you tried to get PKCE working yet? I'm about to give up with Supabase and I'm trying @panva_'s library instead. (I use his code in a bunch of other places, it's pretty high quality stuff IMO.)

Ddave @sathoro Have you tried to get PKCE working yet? I'm about to give up with Supab...

sathoro•5/31/23, 3:11 AM

hmm not sure, we use Google sign-in no problem

Ssathoro hmm not sure, we use Google sign-in no problem

dave•5/31/23, 3:12 AM

Eh I think we’re going to ditch Supabase and do our own OAuth, already done the PoC.

dave•5/31/23, 3:13 AM

It’s the only part of our stack that isn’t multi-region either.

Ddave It’s the only part of our stack that isn’t multi-region either.

sathoro•5/31/23, 3:18 AM

yeah but doesn't really matter since auth code is rarely called compared to everything else

sathoro•5/31/23, 3:18 AM

we allow indefinite logins

Advany•5/31/23, 8:54 AM

I need to do 10k fetches to api end point. Responses can take up to 15 seconds. What would be the cost efficient way to do it? Add them to the queue and process one by one?

Or startup a DO and let it handle it using alarms after each fetch?

What would be the most cost efficient way?

AAdvany I need to do 10k fetches to api end point. Responses can take up to 15 seconds. ...

sathoro•5/31/23, 10:06 AM

using DO as a queue should work well. you would need to do the pricing calculations though. I think we are talking less than a few dollars though (probably much less)...

zegevlier•5/31/23, 10:23 AM

I'm fairly sure 10k requests will fit in the included usage

Ssathoro using DO as a queue should work well. you would need to do the pricing calculati...

Advany•5/31/23, 10:58 AM

I just saw that queues require workers to be on the unbound plan. So that means costs of using queue + workers or DO's would be the same

AAdvany I just saw that queues require workers to be on the unbound plan. So that means ...

sathoro•5/31/23, 10:59 AM

I suggested DOs based on developer experience. we are talking about pennies anyways so I wasn't too worried about cost

Advany•5/31/23, 11:00 AM

yeah, the 10k was only a sample. It will be more like 50 million in production

AAdvany yeah, the 10k was only a sample. It will be more like 50 million in production

sathoro•5/31/23, 11:01 AM

yeah then I would definitely use DOs. Queues is a beta product that has low concurrency and additional latency. with DOs you could create unlimited queues for extra concurrency

sathoro•5/31/23, 11:02 AM

this is what I do in prod. a DO for each user as a queue. works great

Ssathoro yeah then I would definitely use DOs. Queues is a beta product that has low conc...

Advany•5/31/23, 11:02 AM

yeah. I might even go for a non serverless solution. Seems serverless is really expensive for this use case.

AAdvany yeah. I might even go for a non serverless solution. Seems serverless is really ...

sathoro•5/31/23, 11:02 AM

I was thinking about suggesting that since you seemed to be hung up on cost and this is something where a $5 VPS would be the most cost effective thing. just depends on your needs

Ssathoro I was thinking about suggesting that since you seemed to be hung up on cost and ...

Advany•5/31/23, 11:09 AM

yeah. the client needs a cheaper solution in this case... my calculations seem to amount to around $600 for 25 million requests queued and handled

DO seconds 25m * 15 sec = 375 million seconds
DO price = 375 million seconds * 128mb/1gb * $12.50 = $603

+ storage costs + request costs

I think a $100 vps might do a better job handling the use case, unless my calculations are wrong

AAdvany yeah. the client needs a cheaper solution in this case... my calculations seem t...

sathoro•5/31/23, 11:11 AM

that sounds like a lot but that is 4,340 days in processing time lol. that is a heck of a lot of requests especially when they take so long...

Advany•5/31/23, 11:12 AM

yeah. It is... need it to prerender sitemaps... and chrome isn't that fast lol

AAdvany yeah. It is... need it to prerender sitemaps... and chrome isn't that fast lol

sathoro•5/31/23, 11:13 AM

ah...

Ssirupsen At what point of monthly bandwidth to the cache API does someone from Cloudflare...

Hello, I’m Allie!•5/31/23, 11:29 AM

Yeah, CF doesn’t provide an explicit public limit for various reasons. If you are a non-profit/OSS provider, then you might qualify for a discount. Pricing is something you need to contact sales for

Hello, I’m Allie!•5/31/23, 11:29 AM

For anything above PAYGO, it won’t be published

Ssathoro ah...

Advany•5/31/23, 11:57 AM

maybe I can do 4 requests at a time inside a do or worker customer with parallel primoses... didnt think of that. That should lower the cost to $150 a month...

AAdvany maybe I can do 4 requests at a time inside a do or worker customer with parallel...

sathoro•5/31/23, 12:09 PM

yeah you can do like 8 at a time I think

In a Worker, you always have to cache.match and return that response yourself so it’s up to you at w

Similar Threads

Similar Threads

Similar Threads